Pedestrian detection algorithm based on local color parallel similarity features

Abstract

HOG Feature is the mainstream feature applied in the field of pedestrian detection .HOG combined with CSS has good effects on pedestrian detection. Because of the large amount calculation of HOG and CSS, HOG and CSS has poor real-time performance, we propose LCSSF (Local Color Self Similarity Feature) avoiding calculating the global color similarity distribution of CSS. The tested results of the Inria and the street pedestrian database show that the accuracy of the HOG with LCSSF has better detection performance and better real-time performance than HOG and CSS.